Doomed from the Start: Early Abort of LLM Agent Episodes via a Recall-Controlled Probe Cascade

Large language model (LLM) agents solving multi-step tasks frequently commit to trajectories that are doomed to fail, yet continue to consume substantial inference compute before the failure becomes observable. This paper shows that failure is predictable early from the agent's internal representations: lightweight per-round probes on hidden activations anticipate eventual episode failure as early as the first interaction round, where scorers reading only the agent's observable behavior are barely better than chance. The authors turn this signal into a practical abort cascade: one distribution-free calibrated gate per round, with per-round recall budgets jointly searched so that eventually-successful episodes survive all gates at a user-specified global rate. On TextCraft with Qwen-2.5-7B and Llama-3.2-3B, the cascade meets every recall target from 90% to 97% and, at the 90% target, saves 47.1%±10.3% and 37.2%±8.8% of inference compute, 1.6–1.7× the best single-gate policy. An otherwise-identical cascade reading only behavior saves roughly half as much, and adding behavioral features to the probe yields no further gain. Finally, the paper characterizes sample complexity for certifying high recall targets, telling practitioners which recall promises their data can support. Code will be released soon.